Cohomologous Harmonic Cochains

We describe algorithms for finding harmonic cochains, an essential ingredient for solving elliptic partial differential equations in exterior calculus. Harmonic cochains are also useful in computational topology and computer graphics. We focus on finding harmonic cochains cohomologous to a given cocycle. Amongst other things, this allows localization near topological features of interest. We derive a weighted least squares method by proving a discrete Hodge-deRham theorem on the isomorphism between the space of harmonic cochains and cohomology. The solution obtained either satisfies the Whitney form finite element exterior calculus equations or the discrete exterior calculus equations for harmonic cochains, depending on the discrete Hodge star used.

💡 Research Summary

The paper “Cohomologous Harmonic Cochains” presents a thorough study of algorithms for computing harmonic cochains that are cohomologous to a prescribed cocycle. Harmonic cochains—discrete analogues of harmonic differential forms—play a central role in solving elliptic PDEs (e.g., Poisson’s equation) via finite element exterior calculus (FEEC) or discrete exterior calculus (DEC), and they are also useful in computational topology and computer graphics (vector‑field design, conformal parameterization, vortex placement, etc.).

The authors begin by recalling the smooth Hodge‑de Rham theorem, which states that on a closed manifold the space of harmonic p‑forms is isomorphic to the real cohomology group H^p(M;ℝ). For manifolds with boundary they restrict to Neumann (absolute) harmonic fields, which automatically satisfy the natural boundary conditions in the weak formulation. They then discretize the exterior calculus using two frameworks: (1) Whitney‑form FEEC, where the Hodge star is the mass matrix for Whitney forms, and (2) primal‑dual DEC, where the Hodge star is a diagonal matrix defined on primal and dual cells. Both discretizations lead to a discrete Laplace–de Rham operator Δ = d δ + δ d.

Three families of algorithms are examined.

1. Eigenvector‑based methods: By solving the generalized eigenvalue problem Δ u = λ ∗ u, the zero‑eigenvalue eigenspace yields a basis of harmonic cochains. Two formulations are discussed: a direct weak form and a mixed weak form (the latter introduces an auxiliary σ to enforce the divergence‑free condition). Once a basis H is available, any harmonic cochain can be expressed as a linear combination of its columns. This approach is straightforward when the Betti numbers are small, but it requires the expensive eigen‑decomposition of a matrix whose size equals the number of p‑simplices.

2. Projection‑based method: Assuming a pre‑computed harmonic basis H, the given cocycle ω is orthogonally projected onto the harmonic subspace using the inner product induced by the Hodge star. The projection solves the small linear system Hᵀ ∗ H a = Hᵀ ∗ ω for the coefficient vector a, and the desired cohomologous harmonic cochain is h = H a. This method is computationally cheap when H is orthonormal, but it still depends on the prior computation of the full harmonic basis.

3. Weighted least‑squares method (main contribution): The authors prove a discrete analogue of the Hodge‑de Rham theorem: each cohomology class contains a unique harmonic cochain, namely the one of minimal L² norm. For a cocycle ω (i.e., d ω = 0) they formulate the optimization problem
    min_{α∈C^{p‑1}} ‖ω + d α‖_{∗}².
   The first‑order optimality condition yields the normal equation
    dᵀ ∗ d α = −dᵀ ∗ ω.
   Although the matrix dᵀ ∗ d is only positive semidefinite (its kernel corresponds to constants or connected components), the residual d α is uniquely determined, guaranteeing a unique harmonic cochain h = ω + d α. This linear system can be solved directly or with iterative solvers; its sparsity pattern follows that of the coboundary operator d, making it highly scalable. Importantly, because the formulation is a weak form of the Laplace–de Rham operator, Neumann boundary conditions are automatically satisfied, eliminating the need for explicit constraint enforcement.

The paper also discusses the practical impact of the choice of Hodge star. The Whitney mass matrix is accurate for arbitrary simplicial meshes but dense, leading to higher computational cost. The primal‑dual diagonal star is cheap and yields a symmetric positive‑definite system, but may lose accuracy on highly irregular meshes. Numerical experiments on a torus and a multiply‑holed disc demonstrate that both stars produce the same harmonic cochains, confirming the theoretical equivalence.

Finally, the authors provide a constructive proof that the map ϕ: H^p(M) → (H^p(M))^* defined by evaluation on homology cycles is an isomorphism in the discrete setting (for p = 1 in 2‑D meshes). This result underpins the pairing‑based method that combines a homology basis with a harmonic cochain basis to extract cohomologous representatives associated with specific topological features (handles, tunnels, cavities).

In summary, the paper delivers a solid theoretical foundation (discrete Hodge‑de Rham isomorphism), compares three algorithmic strategies, and introduces a simple yet powerful weighted least‑squares technique that avoids pre‑computing harmonic bases while guaranteeing the unique cohomologous harmonic cochain. The work is highly relevant to researchers and practitioners working on PDE solvers, topology‑aware mesh processing, and geometry‑aware graphics applications.